This Article

Citations


Creative Commons License
Except where otherwise noted, this work is licensed under Creative Commons Attribution-NonCommercial 4.0 International License.

Assessing the Effect of Quantitative and Qualitative Predictors on Gastric Cancer Individuals Survival Using Hierarchical Artificial Neural Network Models


1 Department Of Basic Sciences, National Nutrition and Food Technology Research Institute, Faculty of Nutrition Sciences and Food Technology, Shahid Beheshti University of Medical Sciences, Tehran, IR Iran
2 Department of Epidemiology and Biostatistics, School of Public Health, Tehran University of Medical Sciences, Tehran, IR Iran
*Corresponding Author: Department of Epidemiology and Biostatistics, School of Public Health, Tehran University of Medical Sciences, Tehran, IR Iran. Tel.: +98-2188989126, Fax: +98-2188989126, E-mail: zeraatih@tums.ac.ir.
Iranian Red Crescent Medical Journal. 2013 January; 15(1): 42-8. , DOI: 10.5812/ircmj.4122
Article Type: Research Article; Revised: May 26, 2012; Accepted: Jun 11, 2012; epub: Jan 5, 2013; ppub: Jan 5, 2013
Running Title: The Effect of Predictors on Gastric Cancer Patient Survival Using Hierarchical Artificial Neural Network Models

Abstract


Background: There are numerous unanswered questions in the application of artificial neural network models for analysis of survival data. In most studies, independent variables have been studied as qualitative dichotomous variables, and results of using discrete and continuous quantitative, ordinal, or multinomial categorical predictive variables in these models are not well understood in comparison to conventional models.

Objectives: This study was designed and conducted to examine the application of these models in order to determine the survival of gastric cancer patients, in comparison to the Cox proportional hazards model.

Patients and Methods: We studied the postoperative survival of 330 gastric cancer patients who suffered surgery at a surgical unit of the Iran Cancer Institute over a five-year period. Covariates of age, gender, history of substance abuse, cancer site, type of pathology, presence of metastasis, stage, and number of complementary treatments were entered in the models, and survival probabilities were calculated at 6, 12, 18, 24, 36, 48, and 60 months using the Cox proportional hazards and neural network models. We estimated coefficients of the Cox model and the weights in the neural network (with 3, 5, and 7 nodes in the hidden layer) in the training group, and used them to derive predictions in the study group. Predictions with these two methods were compared with those of the Kaplan-Meier product limit estimator as the gold standard. Comparisons were performed with the Friedman and Kruskal-Wallis tests.

Results: Survival probabilities at different times were determined using the Cox proportional hazards and a neural network with three nodes in the hidden layer; the ratios of standard errors with these two methods to the Kaplan-Meier method were 1.1593 and 1.0071, respectively, revealed a significant difference between Cox and Kaplan-Meier (P < 0.05) and no significant difference between Cox and the neural network, and the neural network and the standard (Kaplan-Meier), as well as better accuracy for the neural network (with 3 nodes in the hidden layer). Probabilities of survival were calculated using three neural network models with 3, 5, and 7 nodes in the hidden layer, and it has been observed that none of the predictions was significantly different from results with the Kaplan-Meier method and they appeared more comparable towards the last months (fifth year). However, we observed better accuracy using the neural network with 5 nodes in the hidden layer. Using the Cox proportional hazards and a neural network with 3 nodes in the hidden layer, we found enhanced accuracy with the neural network model.

Conclusions: Neural networks can provide more accurate predictions for survival probabilities compared to the Cox proportional hazards mode, especially now that advances in computer sciences have eliminated limitations associated with complex computations. It is not recommended in order to adding too many hidden layer nodes because sample size related effects can reduce the accuracy. We recommend increasing the number of nodes to a point that increased accuracy continues (decrease in mean standard error), however increasing nodes should cease when a change in this trend is observed.

Keywords: Survival; Life Expectancy; Proportional Hazards Model; Neural Networks

Acknowledgments

None declared.

Footnotes

Implication for health policy/practice/research/medical education Predictions of survival probabilities appear to be more accurate than that achieved with the Cox proportional hazards model, especially currently that sophisticated computations can be conducted. The uses of these new approaches are preferable since they lack the limitations and assumptions associated with the older, more traditional models. For neural networks, we do not recommend adding too many hidden layer nodes since sample-size related effects may reduce the accuracy. We suggest increasing the number of nodes as far as accuracy continues to increase (decrease in mean standard error), however adding nodes should cease when an alteration in this trend is observed.
Please cite this paper as Amiri Z, Mohammad K, Mahmoudi M , Parsaeian M, Zeraati H. Assessing the Effect of Quantitative and Qualitative Predictors on Gastric Cancer Individuals Survival Using Hierarchical Artificial Neural Network Models. Iran Red Cres Med J. 2013;15(1):42-8. DOI:10.5812/ircmj.4122
Financial Disclosure None declared.
Funding/Support None declared.

References


  • 1. Papik K, Molnar B, Schaefer R, Dombovari Z, Tulassay Z, Feher J. Application of neural networks in medicine - a review. Med Sci Monit. 1998;4(3):538-46.
  • 2. Ebell MH. Artificial neural networks for predicting failure to survive following in-hospital cardiopulmonary resuscitation. J Fam Pract. 1993;36(3):297-303. [PubMed]
  • 3. Ashutosh K, Lee H, Mohan CK, Ranka S, Mehrotra K, Alexander C. Prediction criteria for successful weaning from respiratory support: statistical and connectionist analyses. Crit Care Med. 1992;20(9):1295-301. [PubMed]
  • 4. Burke HB. Artificial neural networks for cancer research: outcome prediction. Semin Surg Oncol. 1994;10(1):73-9. [PubMed]
  • 5. Ravdin PM, Clark GM, Hilsenbeck SG, Owens MA, Vendely P, Pandian MR, et al. A demonstration that breast cancer recurrence can be predicted by neural network analysis. Breast Cancer Res Treat. 1992;21(1):47-53. [PubMed]
  • 6. Doyle HR, Dvorchik I, Mitchell S, Marino IR, Ebert FH, McMichael J, et al. Predicting outcomes after liver transplantation. A connectionist approach. Ann Surg. 1994;219(4):408-15. [PubMed]
  • 7. Baxt WG. Use of an artificial neural network for the diagnosis of myocardial infarction. Ann Intern Med. 1991;115(11):843-8. [PubMed]
  • 8. Baxt WG. Analysis of the clinical variables driving decision in an artificial neural network trained to identify the presence of myocardial infarction. Ann Emerg Med. 1992;21(12):1439-44. [PubMed]
  • 9. Ravdin PM, Clark GM. A practical application of neural network analysis for predicting outcome of individual breast cancer patients. Breast Cancer Res Treat. 1992;22(3):285-93. [PubMed]
  • 10. Ohno-Machado L. Sequential use of neural networks for survival prediction in AIDS. Proc AMIA Annu Fall Symp. .170-4. [PubMed]
  • 11. Ripley BD, Ripley RM. Neural networks as statistical methods in survival analysis. In: Dybowski R, Gant V, editor(s). Clinical Applications of Artificial Neural Networks. Cambridge: Cambridge University Press; 2001. p. 237-55.
  • 12. Ripley RM, Harris AL, Tarassenko L. Non-linear survival analysis using neural networks. Stat Med. 2004;23(5):825-42. [DOI] [PubMed]
  • 13. Amiri Z, Mohammad K, Mahmoudi M, Zeraati H, Fotouhi A. Assessment of gastric cancer survival: using an artificial hierarchical neural network. Pak J Biol Sci. 2008;11(8):1076-84. [PubMed]
  • 14. Zeraati H, Mahmoudi M, Kazemnejad A, Mohammed K. Postoperative life expectancy in gastric cancer patients and its associated factors. Saudi Med J. 2005;26(8):1203-7. [PubMed]
  • 15. Biganzoli E, Boracchi P, Coradini D, Grazia Daidone M, Marubini E. Prognosis in node-negative primary breast cancer: a neural network analysis of risk profiles using routinely assessed factors. Ann Oncol. 2003;14(10):1484-93. [PubMed]
  • 16. Jones AS, Taktak AG, Helliwell TR, Fenton JE, Birchall MA, Husband DJ, et al. An artificial neural network improves prediction of observed survival in patients with laryngeal squamous carcinoma. Eur Arch Otorhinolaryngol. 2006;263(6):541-7. [DOI] [PubMed]
  • 17. Suka M, Oeda S, Ichimura T, Yoshida K, Takezawa J. Comparison of proportional hazard model and neural network models in a real data set of intensive care unit patients. Stud Health Technol Inform. 2004;107(Pt 1):741-5. [PubMed]

Table 1.

Distribution of Independent Variables in the Training Group, the Study Group, and the Total Sample

Variable Training Group (n = 165), No. (%) Prediction Group (n = 165), No. (%) Total (n = 330), No. (%) Chi-square, Mean ± SD
Gender 0.06 ± 0.81
Female 52 (31.5) 50 (30.3) 102 (30.9)
Male 113 (68.5) 115 (69.7) 228 (69.1)
History of smoking 0.52 ± 0.47
No 112 (67.9) 118 (71.5) 230 (69.7)
Yes 53 (32.1) 47 (28.5) 100 (30.3)
Pathology 0.06 ± 0.44
Adenocarcinoma 143 (86.7) 138 (83.6) 281 (85.2)
Other 22 (13.3) 27 (16.4) 49 (14.8)
Metastasis 0.45 ± 0.50
Yes 93 (56.4) 99 (40.0) 192 (58.2)
No 72 (43.6) 66 (40.0) 138 (41.8)
T-stage 4.13 ± 0.25
1 10 (6.1) 3 (1.8) 13 (3.9)
2 13 (7.9) 13 (7.9) 26 (7.9)
3 55 (33.3) 54 (32.7) 109 (33.0)
4 87 (52.7) 95 (57.6) 182 (55.2)
N-stage 0.80 ± 0.85
0 104 (63.0) 97 (58.8) 201 (60.9)
1 6 (3.6) 8 (4.8) 14 (4.2)
2 44 (26.7) 47 (28.5) 91 (27.6)
3 11 (6.7) 13 (7.9) 24 (7.3)
M-stage 0.62 ± 0.43
0 149 (90.3) 153 (92.7) 302 (91.5)
1 16 (9.7) 12 (7.3) 28 (8.5)
Number of complementary treatment 5.17 ± 0.13
0 39 (23.6) 28 (17.0) 67 (20.3)
1 42 (25.5) 34 (20.6) 76 (23.0)
2 49 (29.7) 52 (31.5) 101 (30.6)
3 35 (21.2) 51 (30.9) 86 (26.1)
Cancer site 0.50 ± 0.78
Cardia 71 (43.0) 74 (44.8) 145 (43.9)
Antrum 30 (18.2) 33 (20.0) 63 (19.1)
Other 64 (38.8) 58 (35.2) 112 (37.0)
Final status 1.23 ± 0.27
Alive 41 (24.8) 50 (30.3) 91 (27.6)
Deceased 124 (75.2) 115 (69.7) 239 (72.4)
Age, y 65.18 (11.32) 66.04 (10.70) 65.61 (11.01) 0.71 ± 0.48

Figure 1.

Survival probabilities in the training group predicted using Cox and neural network methods (with a 3-node hidden layer) (NN_3) compared to the Kaplan Meier (KM) method (95% limits of agreement)

Figure 2.

Survival probabilities in the study group predicted using three neural network based (NN) methods (with 3-, 5-, and 7-node hidden layers) compared to the Kaplan Meier (KM) method (95% limits of agreement)